Re: Backport of Ingo/Arjan highpte to 2.4.18 (+O1 scheduler)

[Martin J. Bligh]

> First, your backport is clearly broken because it will oops in
> copy_one_pte if the alloc_one_pte_map fails.

That doesn't suprise me ... I did a quick backport without staring
at the code too much, just so I could get some testing number to
see what the difference in performance would be. Arjan is doing
a proper backport to 2.4, and he obviously knows this patch far
better than I, so hopefully he'll address this ;-)

Thanks for pointing this out.

> ....

The bulk of the rest of this will take me a while to think about ;-)

Thanks,

M.

PS. The backport of the 2.5 highpte stuff works fine for me in limited
touch-testing, but I don't have it playing with the discontigmem stuff
yet, so I can't give you any numbers at the moment ...